Introduction
In the simplest model of a telephone speech communication there is a direct, dedicated,
physical connection between the two participants in the conversation, and this link is held for
the duration of the conversation. The analogue electrical signal produced by the telephone at
either end is sent on to connection without modification.
In Pulse Amplitude Modulation (PAM), the unmodified electrical signal is not sent on to the
connection. Instead, short samples of the signal are taken at regular intervals, and these
samples are sent on to the connection. The amplitude of each sample is identical to the signal
voltage at the time when the sample was taken. Typically, 8,000 samples are taken per
second, so that the interval between samples is 125s, and the duration of each sample is
approximately 4s.
Because each sample is very short (~4s) there is a lot of time between samples (~121s).
Samples from other conversations are put into this “spare time”. Usually the samples from 32
separate conversations are put on to a single line. This process is called Time Division
Multiplexing (TDM).
Each sample is very short, and will be distorted as it travels across a communications
network. In order to reconstruct the original analogue signal the only information the receiver
needs to have about a sample is its amplitude, but if this is distorted then all information
about the sample has been lost. To overcome this problem, the pulse is not transmitted
directly, instead its amplitude is measured and converted into an 8 binary number - a
sequence of 1s and 0s. At the receiver end, the receiver merely needs to detect if a 1 or a 0 has
been received so that it can still recover the amplitude of a PAM pulse even if the 1s and 0s
used to describe it have been distorted.
The process of converting the amplitude of each pulse into a stream of 1s and 0s is called
Pulse Code Modulation (PCM)
Note that the process of PAM and PCM (but without the use of TDM) is essentially used to
store music and speech on CDs, but with a higher sample rate, more bits per sample and
complex error correction mechanisms.
Some terms are:
Sampling
The process of measuring the amplitude of a continuous-time signal at discrete
instants. It converts a continuous-time signal to a discrete-time signal.
Quantizing Representing the sampled values of the amplitude by a finite set of levels. It
converts a continuous-amplitude sample to a discrete-amplitude sample.
Encoding Designating each quantized level by a (binary) code.
Sampling and quantizing operations transform an analogue signal to a digital signal.
Use of quantizing and encoding distinguishes PCM from analogue pulse modulation methods.
The quantizing and encoding operations are usually performed in the same circuit at the
transmitter, which is called an Analogue to Digital Converter (ADC). At the receiver end the
decoding operation converts the (8 bit) binary representation of the pulse back into an
analogue voltage in a Digital to Analogue Converter (DAC)
1
28/06/2016
1. Time Division Multiplexing (TDM) - Principle
When sending samples of a signal instead of the signal itself there is time available between
each of the samples. Samples from other analogue signals can be put into this space. The
process of splitting up the
Receiver
Transmitter
time into slots and putting
Timing
Timing
different signals into the
time slots is known as Time
Ch1
Ch1
Buffer
LPF1
i/p
o/p
Division Multiplexing
Transmission Line
Ch1
Ch1
(TDM). A basic real TDM
Buffer
LPF2
i/p
o/p
SW1
SW2
Ch1
Ch1
system interleaves 32
Buffer
LPF3
i/p
o/p
signals and uses electronic
switches. This is a diagram
of a 3 channel PAM-TDM system.
This diagram shows the waveforms produced during the operation of the PAM-TDM system
The switches connect the transmitter and the receiver to each of the channels in turn for a
specific interval of time. In effect each channel is
Ch1
sampled and the sample is transmitted
When the switches are in the channel 1 position,
channel 1 forms a PAM channel with an LPF for
Ch2
reconstruction, and so on for channels 2 and 3. The
result is that the amplitudes samples from each channel
share the line sequentially, becoming interleaved to
Ch3
form a complex PAM wave, as shown above.
A major problems in any TDM system is the
synchronisation of the transmitter and receiver
TDM
timing circuits. The transmitter and receiver must
Signal
switch at the same time and frequency. Also SW1 must
Guard Slots
be in the channel 1 position when SW2 is in the channel
Time
1 position, so that the switches must be synchronised in
Slots 1 2 3 1 2 3 1 2 3
position also.
In a system that uses analogue modulation (PAM) the time slots are separated by guard slots
to prevent crosstalk between channels.
2
28/06/2016
2. Sampling Theorem
Consider a band-limited signal with no frequency components above a certain frequency fm.
The sampling theorem states that this signal can be recovered completely from a set of
samples of its amplitude, if the samples are taken at the rate of fs > 2fm samples per second.
This is often called the uniform sampling theorem for baseband or low-pass signals.
The minimum sampling rate, 2fm samples per second, is called the Nyquist sampling rate (or
Nyquist frequency); its reciprocal l/(2fm) (measured in seconds) is called the Nyquist interval.
fs = 2 * fm is called the Nyquist sampling rate.
For telephone speech the standard sampling rate is 8 kHz (or one sample every 125 s).
2.1 Sampling Methods
Suppose we have an arbitrary signal
(the baseband signal m(t)) which
has a spectrum M(f). Take
infinitesimally short samples of the
Freq
signal m(t) at a uniform rate once
Sampling pulses
fs
2fs
3fs
every ts seconds i.e. at a frequency fs.
This is the ideal form of sampling, it is
Spectrum of Sampling Pulses
Spacing = 1/fs
called instantaneous (or impulse)
sampling.
Samples
fs
2fs
3fs
In effect the signal m(t) is multiplied
by a train of impulses giving rise to a
Time
train of pulses as in the lower line of
Spectrum of Sampled Signal
the diagram. The train of sampling
impulses has a frequency spectrum consisting of all harmonics or multiples of f s and all are at
the same amplitude.
This sampled signal has a spectrum as shown where M(f) is repeated unattenuated
periodically and appears around all multiples of the sampling frequency (fs = 1/ts).
To recover m(t) from the sampled signal we need only pass the sampled signal through a low
pass filter with a stop frequency of fs/2. All of the higher frequency components will be
dropped. In the diagram, if fs is greater than twice the highest frequency in m(t) the
repetitions of the sampled spectra around the harmonics of the fs do not overlap.
Signal(mt)
Spectrum of
Unsampled
Signal M(f)
2.1.1 Flat - top Sampling
An Analogue to Digital Converter requires that the sample value be held constant for a fixed
time until the conversion is completed. This requires a flat-top sampled signal. This has
approximately the same repeated frequency spectrum as with the instantaneous sampling
above, but with each repetition slightly spread out.
The simplest and most common sampling method is performed by a functional block termed a
Sample and Hold (S/H) circuit.
The output from the circuit must be held at a
constant level for the sampling duration. Vcontrol
switches the MOSFET ON until the charge on C
Input signal
is equal to the amplitude of the sampled voltage.
+
Sample
Vcontrol then goes LOW, the MOSFET is OFF
Out
and the charge is held by the capacitor. The
charge held on the capacitor puts a voltage across
Vcontrol
Gnd
the capacitor, and it is held at that value until
the next time that Vcontrol switches the
MOSFET ON. This is called a sample and hold circuit and is usually used as the input to an
ADC.
3
28/06/2016
2.2 Aliasing Error
If a signal is under sampled (sampled at a rate below the
Nyquist rate), the spectrum consists of overlapping
repetitions of the sampled spectrum. Because of the
overlapping tails a single repetition of the spectrum no
Spectrum of Sampled Signal
longer has the complete information about the unsampled
signal, and it is no longer possible to recover it from the
sampled signal. To recover the original signal at the receiving end the sampled signal is
passed through a lowpass filter with a cut off of fs/2, we get a spectrum that is not the
sampled signal but is a different version due to:
 Loss of the tail of the sampled signal spectrum beyond fs/2
 This same tail appears inverted, or folded, onto the spectrum at the cut-off frequency.
This tail inversion is known as aliasing, (or spectral folding or foldover distortion).
The aliasing distortion can be eliminated by cutting the tail (i.e. filtering) of the sampled
signal beyond f > fs/2 before the signal is sampled. By so doing, the overlap of successive
cycles in the sampled signal is avoided. The only error in the recovery of the unsampled signal
is that caused by the missing tail above fs/2.
It is simpler to consider aliasing by considering
fm
fs
2fs
3fs
a single frequency component of m(t). We will
look at the frequency fm and it is sampled at a
rate fs. The diagrams show the frequencies
fs-fm fs+fm 2fs-fm 2fs+fm 3fs-fm 3fs+fm
which will be present in the sampled signal.
There will be frequency components at fm, fs - fm,
fs + fm, 2 fs - fm, 2 fs + fm, 3 fs - fm, 3 fs + fm, etc. etc.
In the first case fm is very much less than fs, so
fm
that fs - fm is much higher than the cut off of the
filter (fs/2).
In the second case fm is below, but close to fs/2,
fs-fm
fs+fm 2fs-fm 2fs+fm 3fs-fm 3fs+fm
so that a sharp cut off filter is required to
ensure that fm is passed but fs - fm is stopped.
In the third case fm is higher than fs/2, so that fs
- fm is less than fs/2. The low pass filter with a
cutoff of fs/2 will therefore block fm (the actual
fm
signal frequency) but will pass a signal with
frequency fs - fm.
This is aliasing
fs-fm
2fs-fm fs+fm 3fs-fm 2fs+fm
3fs+fm
Strictly speaking, a band limited signal does not
exist in reality. It can be shown that if a signal
is time limited it cannot be band limited. All
Frequency
physical signals are necessarily time limited
because they begin at some finite instant and must terminate at some other finite instant.
Hence, all practical signals are theoretically non band limited.
A real signal contains a finite amount of energy, therefore its frequency spectrum must decay
at higher frequencies. Most of the signal energy resides in a finite band, and the spectrum at
higher frequencies contributes little. The error introduced by cutting off the tail beyond a
certain frequency B can be made negligible by making B sufficiently large.
Thus, for all practical purposes a signal can be considered to be essentially band limited at
some value B, the choice of which depends upon the accuracy desired. A practical example of
this is a speech signal. Theoretically, a speech signal, being a finite time signal, has an
infinite bandwidth. But frequency components beyond 3400 Hz contribute a small fraction of
the total energy. When speech signals are transmitted by PCM they are first passed through
a lowpass filter of bandwidth of 3500 Hz. (This filter is called an anti aliasing filter). Higher
sampling rates (i.e. 8000 samples/sec) permits recovery of the signal from its samples using
4
28/06/2016
relatively simple filters i.e. it allows for guard bands between the repetitions of the spectrum
(otherwise recovering signals sampled at the Nyquist rate would require very sharp cut-off
(ideal) filters).
In summary, aliasing distortion produces frequency components in the desired frequency
band that did not exist in the original waveform. Aliasing problems are not confined to speech
digitisation processes. The potential for aliasing is present in any sample data system.
Motion picture taking, for example, is another sampling system that can produce aliasing. A
common example occurs when filming a rotating wheel. Often the sampling process (the
picture refresh rate) is too slow to keep up with the wheel movements and spurious rotational
rates are produced. If the wheel rotates 3550 between frames, it looks to the eye as if it has
moved backwards 50.
3. Pulse Amplitude Modulation
A sampled signal consists of a train of
pulses, where each pulse corresponds
to the amplitude of the signal at the
Input
Transmission
corresponding sampling time. The
Signal Band Limiting
Sampling
Link
Sample
LPF
Filter
Circuit
and Hold
signal sent to line is modulated in
amplitude and hence the name Pulse
Sampling Pulses
Amplitude Modulation (PAM).
A complete PAM system must include a band limiting (or anti aliasing) filter before
sampling to ensure that no spurious or source-related signals get folded back into the desired
signal bandwidth - no aliasing. The input filter may also be designed to cut off very low
frequencies to removed 50 Hz hum from power lines.
Several PAM signals can be multiplexed together as long as they are kept distinct and are
recoverable at the receiving end. This system is one example of Time Division Multiplex
(TDM) transmission (although it has never been widely used fo speech, it has applications in
remote monitoring and telemetry).
The sample-and-hold circuit takes in each pulse and holds the amplitude of that pulse until
the arrival of the next pulse. It produces a staircase approximation to the sampled wave form.
With use of the staircase approximation, the power level of the signal coming out of the
reconstructive filter (LPF) is nearly the same as the level of the sampled input signal.
The filters are assumed to have ideal characteristics - not like real filters. Filters with real
attenuation slopes at the band edge can be used if the input signal is slightly over sampled. If
sampling frequency is greater than twice the bandwidth, the spectral bands are sufficiently
separated from each other that filters with gradual roll-off characteristics can be used.
As an example, sampled voice systems typically use band limiting filters with a 3 dB cut-off
around 3.4 kHz and a sampling rate of 8 kHz. Thus the sampled signal is sufficiently
attenuated at of 4 kHz to adequately reduce the energy level of the foldover spectrum.
5
28/06/2016
4. Pulse Code Modulation
Pulse Code Modulation (PCM) is an extension of PAM wherein each analogue sample value is
quantized into a discrete value for representation as a digital code word.
Thus, as shown below, a PAM system can be converted into a PCM system by adding a
suitable analogue-to-digital (A/D) converter at the source and a digital-to-analogue (D/A)
converter at the destination.
PCM is a true digital process
Modulator
as compared to PAM. In
Analogue
PCM
PCM the speech signal is
Parallel
Digital
Input
Output
A to D
Binary
Sampler
to Serial
Pulse
converted from analogue to
Converter
Coder
Converter
Generator
digital form.
PCM is standardised for
Demodulator
telephony by the ITU-T
PCM
Analogue
(International
Serial to
Input
D to A
Output
LPF
Parallel
Telecommunications Union Converter
Converter
Telecoms, a branch of the
UN), in a series of
recommendations called the G series. For example the ITU-T recommendations for out-ofband signal rejection in PCM voice coders require that 14 dB of attenuation is provided at 4
kHz. Also, the ITU-T transmission quality specification for telephony terminals require that
the frequency response of the handset microphone has a sharp roll-off from 3.4 kHz.
In quantization the levels are assigned a binary codeword. All sample values falling between
two quantization levels are considered to be located at the centre of the quantization interval.
In this manner the quantization process introduces a certain amount of error or distortion
into the signal samples. This error known as quantization noise, is minimised by establishing
a large number of small quantization intervals. Of course, as the number of quantization
intervals increase, so must the number or bits increase to uniquely identify the quantization
intervals. For example, if an analogue voltage level is to be converted to a digital system with
8 discrete levels or quantization steps three bits are required. In the ITU-T version there are
256 quantization steps, 128 positive and 128 negative, requiring 8 bits. A positive level is
represented by having bit 8 (MSB) at 0, and for a negative level the MSB is 1.
4.1 Quantization
This is the process of setting the sample amplitude, which can be continuously variable to a
discrete value. Look at Uniform Quantization first, where the discrete values are evenly
spaced.
Output
4.1.1 Uniform Quantization
-mp
+mp

Input
We assume that the amplitude of the signal m(t) is confined
to the range (-mp, +mp ). This range (2mp) is divided into L
levels, each of step size , given by
 = 2 mp / L
A sample amplitude value is approximated by the midpoint
of the interval in which it lies. The input/output
characteristics of a uniform quantizer is shown.
6
28/06/2016
4.2 Companding
In a uniform or linear PCM system the size of every quantization interval is determined by
the SQR requirement of the lowest signal to be encoded. This interval is also for the largest
signal - which therefore has a much better SQR.
Example:
A 26 dB SQR for small signals and a 30 dB dynamic range produces a 56
dB SQR for the maximum amplitude signal.
In this way a uniform PCM system provides unneeded quality for large signals. In speech the
max amplitude signals are the least likely to occur. The code space in a uniform PCM system
is very inefficiently utilised.
A more efficient coding is achieved if the quantization intervals increase with the sample
value. When the quantization interval is directly proportional to the sample value ( assign
small quantization intervals to small signals and large intervals to large signals) the SQR is
constant for all signal levels. With this technique fewer bits per sample are required to
provide a specified SQR for small signals and an adequate dynamic range for large signals
(but still with the SQR as for the small signals). The quantization intervals are not constant
and there will be a non linear relationship between the code words and the values they
represent.
compressed signal level
Originally to produce the non linear
quantization the baseband signal was
passed through a non-linear amplifier with
input/output characteristics as shown
input signal level
before the samples were taken. Low level
signals were amplified and high level
signals were attenuated. The larger the
sample value the more it is compressed
before encoding. The PCM decoder expands the compressed value using an inverse
compression characteristic to recover the original sample value. The two processes are called
companding.
There are 2 companding schemes to describe the curve above:
1. -Law Companding (also called log-PCM)
This is used in North America and Japan. It uses a logarithmic compression curve which is
ideal in the sense that quantization intervals and hence quantization noise is directly
proportional to signal level (and so a constant SQR).
2. A- Law Companding
This is the ITU-T standard. It is used in Europe and most of the rest of the world. It is very
similar to the -Law coding. It is represented by straight line segments to facilitate digital
companding.
Originally the non linear function was obtained using non linear devices such as special
diodes. These days in a PCM system the A to D and D to A converters (ADC and DAC) include
a companding function.
4.3 PCM Encoding Process (HDB3)
The output from the analogue to digital converter (ADC) has n parallel bits. In the case of
telephony n = 8. The most significant bit is the signed bit. If the measured sample is positive
then the signed bit is 0. If the measured sample is negative then the signed bit is 1. The
remaining 7 bits are used to code the sample value. The ITU-T define a look up table which
allocates a particular binary code to each quantified A-law value.
The line coding which is used assigns opposite polarities to successive “1”s. This eliminates
any DC voltage on the line, and reduces the inter symbol interference if adjacent bits are “1”.
7
28/06/2016
If there is silence on the PCM channel then the measured samples will be 0 Vrms and the
output of the DAC will be 1000 0000. A stream of all zeros is not desirable on an active
channel because
 all zeros could also be a fault condition and
 it is difficult to recover the clock signal from the incoming signal.
The coding system HDB3 is used and was developed to eliminate all zeros, and to assign
opposite polarities to successive “1”s.
This is a bipolar signalling technique (i.e. relies on the transmission of both positive and
negative pulses).
In AMI positive and negative pulses (of equal amplitude) are used for alternative symbols 1.
No pulse is used for symbol 0. In either case the pulse returns to 0 before the end of the bit
interval. This eliminates any DC on the line.
HDB3 encoding rules follow those for AMI, except that a sequence of four consecutive 0's are
encoded using a special "violation" bit. The 4th 0 bit is given the same polarity as the last 1-bit
which was sent using the AMI encoding rule. This prevents long runs of 0's in the data stream
which may otherwise prevent a receiver from tracking the centre of each bit. By introducing
violations, extra "edges" are introduced, enabling a Digital PLL to reliably reconstruct the
clock signal at the receiver. The HDB3 is transparent to the sequence of bits being
transmitted (i.e. whatever data is sent, the Digital PLL can reconstruct the data and extract
the bits at the receiver).
To prevent a DC being introduced by excessive runs of zeros any run of more than four zeros
encodes as B00V. The value of B is assigned + or - alternately throughout the bit stream.
Example
1111 1111
=
+ - + - + - + B B B B B B B B
1010 1010
=
+ 0 - 0 + 0 - 0
B 0 B 0 B 0 B 0
1000 0001
1000 0110
=
+ 0 0 0 + 00 B 0 0 0 V 00B
=
=
+ 0 0 0 + - +0
B 0 0 0 V BB0
4.4 PCM Timing and Synchronisation
The PCM receiver must be able to identify the start and finish of each full sampling sequence
and to identify each bit position. The sampling clock needs to be either sent to, or regenerated
at, the receiving side to determine when each full sequence of sampling begins and ends. The
data clock is also needed to determine exactly when to read each bit of information.
A PCM channel is sampled at 8,000 Hz or once every 125 s. If there is one channel or 30
TDM channels the sampling period is fixed at 125
data clock 64 kbit/s
15.625 s
s and this period is known as a frame. Therefore
the frame clock must have a period of 125 s. The
rising edge of the frame clock informs the receiver
frame clock 125 s
that the next bit will be Bit 1 of a new sample. The
falling edge of the data clock informs the receiver
that it must read the data bit.
When the bit stream is transmitted along a line
B1 B2 B3 B4 B5 B6 B7 B8 B1
the pulses become distorted and the rise and fall
1
0
1
0
0
1
1
1
?
times become significant. Ideally, a 1 will be “high”
8
28/06/2016
for 15.625 s. In practice the pulse may only be above the “high” threshold for a few s so it is
very important that the bit is read within a certain time limit of the clock pulse.
The simplest way to synchronise a PCM sender to a PCM receiver is to send the clock signals
on different circuits to the data This would be done in a self-contained system such as private
branch exchange (PBX). Telephony is full duplex so that there is a coder and a decoder at
each port, but each would use the same clock.
To minimise the number of circuits it is possible to use a line-coding scheme which allows the
receiver to extract the clocks from the PCM signal. In this case the receiver will have free
running clocks that lock (using a PLL) to the phase and frequency of the transitions in the
data stream. The line-coding scheme ensures that there is a transition for every data bit.
9
28/06/2016
5. Differential pulse coding schemes
PCM transmits the absolute value of the signal for each frame. Instead we can transmit
information about the difference between each sample. The two main differential coding
schemes are:

Delta Modulation

Differential PCM and Adaptive Differential PCM (ADPCM)
5.1 Delta Modulation
Delta modulation converts an analogue signal, normally voice, into a digital signal.
The analogue signal is sampled as in the PCM
granular noise
process. Then the sample is compared with
1
1
the previous sample. The result of the
1
comparison is quantified using a one bit coder.
1
0
0
1
If the sample is greater than the previous
0
1
1
sample a 1 is generated. Otherwise a 0 is
0
0
1
0
0
0
generated. The advantage of delta modulation
0
over PCM is its simplicity and lower cost. But
the noise performance is not as good as PCM.
To reconstruct the original from the quantization, if a 1 is received the signal is increased by a
step of size q, if a 0 is received the output is reduced by the same size step. Slope overload
occurs when the encoded waveform is more than a step size away from the input signal. This
condition happens when the rate of change of the input exceeds the maximum change that
can be generated by the output. Overload will occur if:
dx(t)/dt  q /T = q * fs
where: x(t) = input signal, q = step size, T = period between samples, f s = sampling frequency
Assume that the input signal has maximum amplitude A and maximum frequency F. The
most rapidly changing input is provided by x(t) = A * sin (2 *  * F * t).
For this
dx(t)/dt = 2 *  * F * A * sin (2 *  * F * t).
This slope has a maximum value of
2**F*A
Overload occurs if
2 *  * F * A > q * fs
To prevent overload we require
q * fs > 2 *  * F * A
Example A = 2 V, F = 3.4 kHz, and the signal is sampled 1,000,000 times per second,
requires q > 2 * 3.14 * 3,400 * 2 /1,000,000 V > 42.7 mV
Granular noise occurs if the slope changes more slowly than the step size. The reconstructed
signal oscillates by 1 step size in every sample. It can be reduced by decreasing the step size.
This requires that the sample rate be increased. Delta Modulation requires a sampling rate
much higher than twice the bandwidth. It requires oversampling in order to obtain an
accurate prediction of the next input, since each encoded sample contains a relatively small
amount of information. Delta Modulation requires higher sampling rates than PCM.
5.2 Differential PCM (DPCM) and ADPCM
Differentiator
Analogue
Input Band Limiting
Filter
+
-
Quantiser
Enoder
ADC
Encoded
Difference
Samples
Accumulator
DAC
DPCM is also designed to take advantage
of the redundancies in a typical speech
waveform. In DPCM the differences
between samples are quantized with fewer
bits that would be used for quantizing an
individual amplitude sample. The
sampling rate is often the same as for a
comparable PCM system, unlike Delta
Modulation.
Adaptive Differential Pulse Code Modulation ADPCM is standardised by ITU-T
recommendations G.721 and G.726. The method uses 32,000 bits/s per voice channel, as
10
28/06/2016
compared to standard PCM’s 64,000 bits/s. Four bits are used to describe each sample, which
represents the difference between two adjacent samples. Sampling is 8,000 times a second. It
makes it possible to reduce the bit flow by half while maintaining an acceptable quality.
While the use of ADPCM (rather than PCM) is imperceptible to humans, it can significantly
reduce the throughput of high speed modems and fax transmissions.
The principle of ADPCM is to use our knowledge of the signal in the past time to predict the
signal one sample period later, in the future. The predicted signal is then compared with the
actual signal. The difference between these is the signal which is sent to line - it is the error
in the prediction. However this is not done by making comparisons on the incoming audio
signal - the comparisons are done after PCM coding.
To implement ADPCM the original (audio) signal is sampled as for PCM to produce a code
word. This code word is manipulated to produce the predicted code word for the next sample.
The new predicted code word is compared with the code word of the second sample. The result
of this comparison is sent to line. Therefore we need to perform PCM before ADPCM.
The ADPCM word represents the prediction error of the signal, and has no significance itself.
Instead the decoder must be able to predict the voltage of the recovered signal from the
previous samples received, and then determine the actual value of the recovered signal from
this prediction and the error signal, and then to reconstruct the original waveform.
ADPCM is sometimes used by telecom operators to fit two speech channels onto a single
64 kbit/s link. This was very common for transatlantic phone calls via satellite up until a few
years ago. Now, nearly all calls use fibre optic channels at 64 kbit/s.
11
28/06/2016